Knowledge catalogue
SemiAnalysis: A detailed look at Jalapeño, OpenAI's ASIC developed with Broadcom in 16 months, which beat Nvidia, AMD, and Google chips on multiple top open-weight models — OpenAI's self-designed ASIC
OpenAI reports Jalapeño delivers 1.5–1.9× more AI work per watt and 1.7–3.6× lower latency vs comparison systems on InferenceX.
OpenAI’s Jalapeño is a self‑designed ASIC for large‑language‑model inference that was taped‑out in about 16 months, developed with Broadcom and featuring HBM4 memory. Benchmarks on the InferenceX suit
OpenAI Jalapeno is spicy Usually first generation chips aren't competitive, but OpenAI is beating Nvidia Blackwell and even Rubin This is huge news! We got to go into OpenAI's lab to dissect their new
Edward Ludlow / Bloomberg: Waymo says it has built an ASIC chip that will improve its robotaxis' reflexes and navigational skills and help it diversify away from companies like Nvidia — Alphabet Inc.'
Ian King / Bloomberg: Sources: Qualcomm reached a deal with ByteDance to supply millions of ASICs for AI data centers to support AI agents in the Doubao chatbot; QCOM jumps 5%+ — Qualcomm Inc. reached
arXiv:2512.05402v2 Announce Type: replace-cross Abstract: Bitcoin mining hardware acquisition requires strategic timing due to volatile markets, rapid technological obsolescence, and protocol-driven r
@bookwormengr: A look at DeepSeek's model optimization to reduce HBM use, potentially enabling domestic memory, ASIC, and CPU makers to create a Chinese AI hardware ecosystem — Have you ever wondered,
Dwarkesh Patel / Dwarkesh Podcast: Q&A with Jensen Huang on Nvidia's supply chain moat, competition from ASICs like Google's TPU, investing in AI labs and neoclouds, selling to China, and more — “If o
arXiv:2604.17808v1 Announce Type: cross Abstract: Zero-knowledge proof (ZKP) provers remain costly because multi-scalar multiplication (MSM) and number-theoretic transforms (NTTs) dominate runtime as
Hi i'm dwarkesh! Grew up all over the US, now sf-based and always down to nerd out about AI, science & history :) a lil about me: 🟠 Host of the dwarkesh podcast 🟠 Studied at UT Austin 🟠 Just published
arXiv:2512.24086v2 Announce Type: replace Abstract: In video and image generation tasks, Diffusion Transformer (DiT) models incur extremely high computational costs due to attention mechanisms, which
SAUVER LA FRANCE ET L'EUROPE EN FAISANT LE SEUL PARI QUI VAILLE (série banane rouge) I. COMME BONAPARTE : CONCENTRER LES FORCES, PAS LES DILUER Dans le discours politico-industriel d'aujourd'hui, on e
arXiv:2406.06543v2 Announce Type: replace-cross Abstract: Deep learning has driven significant technological advancements, but its high energy consumption limits its use on battery-operated edge devic
Most popular model on @OpenRouter (10tr tokens) turns out to be a 1.6tr MoE by @Meituan_LongCat (superapp/DoorDash of China) Basically Gemini / Opus 4.6 level 35tr tokens trained entirely on 50k Chine
Nvidia would likely experience downward pressure on pricing power and profit margins if an AI‑chip glut materializes, as excess supply from competitors (e.g., ASICs or lower‑cost GPUs) could force the
I discovered recently that Google uses their own TPUs, like tiny ASIC cards like the toy ones that existed for bitcoin. And while it sounds inefficient the fact they use thousands of them because...th
Auto-generated index of all companies mentioned across the wiki.
Dominic-Madori Davis / TechCrunch: Ghost, which makes a 3,499 computer designed for AI agents and includes an RTX Pro 4000 SFF Blackwell GPU, emerges from stealth with an 11M seed led by a16z — For mu
Dr. Ian Cutress / More Than Moore: Q&A with OpenAI VP of Hardware Richard Ho on its Jalapeño inference chip co-designed with Broadcom, using internal OpenAI models to design the chip, and more — VP Ha
I've been wanting to test out LLM inference on FPGAs for a while now, but didn't have a big reason too since there were no frontier class models at 9B-27B scale (3.6 27B was great still but it didn't
two weeks ago i went on @swyx's pod and said some things that i... should not have said. a lot has happened since then, i owe you all an apology. i'm sorry that i was right about every single thing. a
arXiv:2606.06159v1 Announce Type: cross Abstract: Spiking neural networks (SNNs) have the potential to emerge as the third generation of neural networks and have attracted increasing attention across
arXiv:2604.13543v1 Announce Type: cross Abstract: Long Short-Term Memory (LSTM) neural networks have penetrated healthcare applications where real-time requirements and edge computing capabilities are
Completely agree with Jensen! Not exporting AI out of fear is classic case where the cure would be 100x worse than the disease. You slow down innovation, progress and US technology and economic leader
The Information: Sources: OpenAI agrees to pay Cerebras $20B+ to use its server chips, double the amount previously associated with the deal, and may receive equity in Cerebras — OpenAI recently struc
Dylan Patel of SemiAnalysis posted a clarification on X distinguishing NVIDIA from automotive companies, likely in response to comparisons or confusion surrounding NVIDIA's growing presence in the aut
arXiv:2508.12021v2 Announce Type: replace Abstract: Unsupervised federated learning (UFL) enables privacy-preserving distributed training without data labeling, yet practical deployment remains challe
arXiv:2609.38477v1 Announce Type: cross Abstract: Large language models (LLMs) incur substantial storage, memory-bandwidth and energy costs, motivating compact weight representations. Existing seed-ba
arXiv:2609.32930v1 Announce Type: cross Abstract: Brain-computer interfaces (BCIs) have emerged as transformative technologies that enable direct communication between the brain and external devices.
Kai Nicol-Schwarz / CNBC: Jensen Huang says AI model distillation is “competition”; Scott Bessent described it as “theft” in July and threatened sanctions against overseas companies — Watch CNBC's ful
arXiv:2609.13947v1 Announce Type: cross Abstract: In-sensor computing reduces the cost of transmitting high-resolution image data by performing early-stage processing near the sensor. However, the log
arXiv:2609.00450v1 Announce Type: cross Abstract: Block Quantization (BQ) is a promising approach for efficient deployment of large language models (LLMs), enabling low-precision computation with cont
Anna Tong / Forbes: Docs: US data labeling companies, like Surge AI and Mercor, that sell training datasets to US AI labs and the government are also selling them to Chinese labs — The same Silicon Va
There is a lot of buzz around running 100B parameter models on cheap local hardware using ternary (1.58-bit) quantization like Microsoft's BitNet architecture. The theoretical hardware shortcuts are i
My 2hr workshop on Open vs Closed models, reward hacking, benchmaxxing & RL is out! 1. Closed vs open models 2. Throughput maxxing but accuracy minimizing 3. Benchmaxxing & cheating 4. Distillation &
arXiv:2602.07400v2 Announce Type: replace Abstract: Gradient-based LUT- and logic-gate-based neural networks (LUTNet, LogicNets, DiffLogic, PolyLUT, NeuraLUT, WARP-LUT, DWN, LILogicNet, LightLUT) repl
arXiv:2607.05590v1 Announce Type: cross Abstract: Implantable brain-computer interfaces require on-node spike sorting to reduce telemetry bandwidth and power while maintaining reliable neural decoding
OpenAI has just revealed a new 'intelligence processor' chip for AI servers made in partnership with Broadcom. The chip, called Jalapeño, is designed to power current and future large language models,
arXiv:2606.09946v1 Announce Type: cross Abstract: Edge-AI systems increasingly require real-time CNN inference under strict energy, performance, security, and privacy constraints. Approximate computin
arXiv:2606.03940v1 Announce Type: cross Abstract: In robotics systems, vast amounts of visual data are easily captured at high resolution using low-cost, low-power hardware. Yet, limited bandwidth and
arXiv:2602.08916v2 Announce Type: replace-cross Abstract: Objective: Acute mountain sickness (AMS) is the most prevalent altitude illness, affecting unacclimatized individuals ascending above 2,500 m
arXiv:2605.06878v1 Announce Type: cross Abstract: This paper presents CARMEN, a runtime-adaptive, CORDIC-accelerated multi-precision vector engine for resource-efficient deep learning inference. The k
Dylan Patel discusses the advantages of drones utilizing American technology infrastructure and components, likely highlighting benefits such as reliability, security, supply chain independence, and t
Benjamin Wallace / New York Times: A profile of Dwarkesh Patel, whose podcast has become mandatory listening in the AI community, with guests like Jensen Huang, Elon Musk, and Mark Zuckerberg — Dwarke
Eli Rosenberg / The Information: 33% of S&P 500 companies provided security perks for execs in 2025: Jensen Huang's security cost rose from 690K in 2023 to 3.5M in 2025, Zuckerberg spent $22M — Nvidia
Jordan Novet / CNBC: Nvidia stock jumps 4.3% to close at a record for the first time since Oct., pushing Nvidia's market cap past 5T, as a rally in Intel pushed chipmakers higher — Nvidia shares close
Alexandra Alper / Reuters: US Commerce Secretary Howard Lutnick says Nvidia has yet to sell H200 chips to Chinese companies and that the Chinese government has not approved such purchases — Nvidia's (
Sam Altman posted about OpenAI's successful pilot program with NVIDIA to deploy Codex (an AI code generation model) across an entire company, highlighting positive results from the enterprise rollout.
Rebecca Bellan / TechCrunch: Source: Mira Murati's TML signed a deal with Google Cloud, valued in single-digit billions, to access Google's latest AI systems built on Nvidia's GB300 chips — Former Ope
The Korea Herald: SK hynix says it has begun mass production of the 192GB SOCAMM2, a next-gen LPDDR5X low-power DRAM module designed particularly for Nvidia's Vera Rubin — SK hynix Inc. said Monday it
Abram Brown / The Information: A look at Dylan Patel's SemiAnalysis, an AI newsletter and research firm that expects $100M+ in 2026 revenue from subscriptions and AI supply chain research — For Dylan
Qianer Liu / The Information: Sources: Google is in talks with Marvell Technology to develop a memory processing unit that works alongside TPUs, and a new TPU for running AI models — Google is in talk
1/ Old Nvidia chips are getting MORE valuable with age, not less. That has never happened in computing history. Marc Andreessen (@pmarca) on @latentspacepod with Alessio Fanelli (@FanaHOVA) and @swyx
Zvi Mowshowitz / Don't Worry About the Vase: A deep dive into Dwarkesh Patel's interview with Jensen Huang, including Huang's takes on Nvidia's moat and chip sales to China, and reactions to the inter